A multi-layer invisible information printing method and detection system for cultural and creative packaging

Abstract

The embodiment of the application discloses a multilayer invisible information printing method and detection system for cultural and creative packaging. The method comprises the following steps: obtaining target visible image data for cultural and creative packaging and at least two layers of invisible information data to be embedded; constructing a visible consistency constraint model based on the target visible image data; constructing an invisible separability constraint model for each layer of invisible information data; under the condition of simultaneously satisfying the visible consistency constraint model and each invisible separability constraint model, performing joint optimization calculation on the target visible image data and each layer of invisible information data to generate single printing data; and using the single printing data for printing output of the cultural and creative packaging. The application realizes invisible information embedding and detection which is visually consistent under normal observation and layer-by-layer distinguishable under multiple preset identification conditions.

Description

Technical Field

[0001] This application relates to the fields of digital image processing and information hiding technology, and in particular to a method and detection system for printing multi-layer invisible information for cultural and creative packaging. Background Technology

[0002] With the development of the cultural and creative industries, cultural and creative packaging not only needs to possess aesthetic and protective functions, but also increasingly emphasizes interactivity, anti-counterfeiting, and cultural narrative functions. In existing technologies, the concealment of invisible information in printed materials mainly employs digital watermarking or steganography methods, such as embedding based on Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT), or using ultraviolet / infrared inks and polarizing materials to achieve physical layer separation. These methods are effective against single-layer or simple texture backgrounds, but in the highly artistic and complex textured images (such as traditional Chinese paintings, paper-cutting, and ink wash patterns) of cultural and creative packaging, they face the following problems: embedded information is easily interfered with by visible image textures, leading to visual distortion or low capacity; there is a lack of effective separation mechanisms between multiple layers of information, resulting in severe interlayer interference after printing distortion (such as ink dot diffusion and registration deviation); the embedding and detection processes are separated, lacking a unified constraint model, leading to poor detection robustness.

[0003] Therefore, there is an urgent need for a joint optimization method that considers perceptual consistency and interlayer separability to achieve the generation and reliable detection of single print data. Summary of the Invention

[0004] This application provides a method and detection system for printing and detecting multi-layered invisible information in cultural and creative packaging. By constructing a visible consistency constraint model and an invisible separability constraint model, and generating single printing data under a joint optimization framework, it achieves the embedding and detection of invisible information that is visually consistent under normal observation and distinguishable layer by layer under multiple preset recognition conditions.

[0005] This application provides the following solution:

[0006] According to a first aspect, a method for printing multi-layer invisible information for cultural and creative packaging is provided. The method includes: acquiring target visible image data for cultural and creative packaging, and at least two layers of invisible information data to be embedded; constructing a visible consistency constraint model based on the target visible image data, wherein the visible consistency constraint model is used to limit the generated image to maintain consistency with the target visible image at the visual perception level under normal observation conditions; constructing invisible separability constraint models for each layer of invisible information data, wherein the invisible separability constraint models are used to limit the layer of invisible information to have distinguishable decoding features under preset recognition conditions; and performing joint optimization calculations on the target visible image data and each layer of invisible information data while simultaneously satisfying the visible consistency constraint model and each of the invisible separability constraint models to generate single printing data, and using the single printing data for printing output of cultural and creative packaging.

[0007] According to one achievable method in an embodiment of this application, the step of constructing a visibility consistency constraint model based on the target visible image data includes: performing perceptual feature decomposition on the target visible image data to extract a set of structural features related to human visual stability; determining a visible stability interval that maintains visual perception stability under different observation scales and illumination conditions based on the set of structural features; mapping the visible stability interval to constraint feature description parameters, and using the constraint feature description parameters to limit the range of image changes allowed to occur during the embedding of latent information, thereby forming the visibility consistency constraint model.

[0008] According to one of the embodiments of this application, the preset identification conditions include at least one or more of the following: illumination band, imaging angle, imaging resolution, image acquisition device type, or image preprocessing parameters.

[0009] According to one achievable method in the embodiments of this application, the step of constructing a stealth separability constraint model for each layer of stealth information data includes: establishing a corresponding set of identification condition parameters for each layer of stealth information data, and extracting hierarchical discrimination features of the stealth information based on the set of identification condition parameters; performing distribution modeling on the hierarchical discrimination features of each layer of stealth information in a unified discrimination feature space to obtain the discrimination region corresponding to each layer of stealth information; and limiting the minimum distinguishing interval of each layer of stealth information in the discrimination feature space based on the interrelationship between the discrimination regions, and mapping the minimum distinguishing interval to the constraint conditions of the stealth information embedding stage, thereby forming the stealth separability constraint model.

[0010] According to one achievable method in an embodiment of this application, the step of defining the minimum distinguishing interval of each layer of hidden information in the discriminative feature space based on the interrelationship between the discriminative regions includes: calculating the degree of overlap between each layer of discriminative regions; and adaptively adjusting the minimum distinguishing interval based on the degree of overlap, using a preset initial minimum distinguishing interval as a basis. The adaptive adjustment includes: if the degree of overlap exceeds a preset threshold, increasing the minimum distinguishing interval of each relevant layer proportionally until the degree of overlap is lower than the threshold or reaches a preset upper limit; if the degree of overlap is lower than the threshold, reducing the minimum distinguishing interval proportionally to improve the embedding capacity, while ensuring that the preset minimum security interval requirement is not violated.

[0011] According to one achievable method in an embodiment of this application, the step of performing joint optimization calculation on the target visible image data and the latent information data of each layer to generate single printed data includes: constructing a joint representation structure for describing the target visible image data and the latent information data of each layer within a unified data representation space; applying collaborative constraints to the joint representation structure based on the visible consistency constraint model and each of the latent separability constraint models to form multi-constraint-restricted generation conditions; and under the generation conditions, iteratively updating the representation state of each information component in the joint representation structure, and outputting the corresponding single printed data when the convergence judgment conditions of each constraint model are met.

[0012] According to one achievable method in an embodiment of this application, the step of iteratively updating the representation state of each information component in the joint representation structure includes: iteratively optimizing the joint representation structure using gradient descent; in each iteration step, simultaneously calculating a visible consistency loss term and a latent separability loss term; the visible consistency loss term includes at least a weighted combination of a perceptual loss function and a structural similarity measure; the latent separability loss term includes a weighted combination of a minimum inter-layer distance constraint and a minimum mutual information constraint in the discriminative feature space; when the weighted sum of the visible consistency loss term and the latent separability loss term is lower than a preset convergence threshold, or when the maximum number of iterations is reached, the optimization is determined to have converged and a single printed data is output.

[0013] According to the second aspect, a multi-layer invisible information detection system for cultural and creative packaging is provided. The system is used to detect printed images formed after single printing data generated in any one of the first aspects is printed on cultural and creative packaging. The system includes: a data acquisition module for acquiring detection image data of the printed image; a perception condition modeling module for modeling the perception feature space corresponding to the detection image data according to preset recognition conditions; a hierarchical detection control module for determining the current invisible information level to be detected based on the perception feature space and pre-established hierarchical detection rules; an invisible information decoding module for performing targeted decoding calculations on the detection image data under the determined invisible information level to obtain candidate invisible information for the corresponding level; a consistency and confidence judgment module for verifying the consistency between the candidate invisible information and the invisible information constraint features generated during the printing stage, and calculating the corresponding decoding confidence; and a detection result generation module for judging the detection results of each layer of invisible information based on the consistency and confidence.

[0014] According to one achievable method in an embodiment of this application, the perceptual condition modeling module is configured to: perform feature extraction processing on the detected image data under the preset recognition conditions in a manner that is adapted to or corresponds to the perceptual feature decomposition method in the visible consistency constraint model described in the second item of the first aspect above, so as to construct a perceptual feature space for hierarchical detection.

[0015] According to one achievable method in this application embodiment, the step of determining the detection results of each layer of hidden information based on the consistency and confidence level includes: filtering each layer of hidden information and its corresponding decoding confidence level, and eliminating candidate information with a confidence level lower than a preset threshold; weighting and fusing the remaining layers of hidden information according to their confidence levels to generate a comprehensive detection result; mapping the comprehensive detection result into a multi-layer visualization output, including the display position, decoding status, and confidence level index of each layer of hidden information; recording the decoding conditions, decoding time, and confidence level for each output information to form a traceable detection log; and outputting the visualization output and the detection log simultaneously for use by the user or subsequent system calls.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0017] This application constructs a visible consistency constraint model and a hidden separability constraint model, and performs joint optimization calculations on the target visible image and multi-layered hidden information under dual constraints to generate single printed data. This achieves the following: after the cultural and creative packaging is printed, it maintains a high degree of visual consistency with the original visible image under normal observation conditions, and reliably decodes the corresponding layer-by-layer hidden information under different preset recognition conditions. This method effectively solves the traditional problem of embedding multi-layered hidden information in the highly artistic and complex texture background of cultural and creative packaging. It ensures extremely high imperceptibility and aesthetics while significantly improving inter-layer distinguishability and detection robustness, avoiding interference and capacity loss caused by independent layer-by-layer embedding. It is applicable to interactive anti-counterfeiting and narrative enhancement of cultural and creative products such as tea, books, and intangible cultural heritage derivatives, and has outstanding practical value and innovation.

[0018] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a method for printing multi-layer invisible information on cultural and creative packaging, provided as an embodiment of this application;

[0021] Figure 2 This is a structural block diagram of a multi-layer invisible information detection system for cultural and creative packaging, provided as an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0023] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0025] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0026] Figure 1 This is a flowchart illustrating a method for printing multi-layer invisible information on cultural and creative packaging, as provided in an embodiment of this application. Figure 1 As shown, the method may include the following steps:

[0027] Step 101: Obtain the target visible image data for cultural and creative packaging, as well as at least two layers of hidden information data to be embedded.

[0028] Step 102: Based on the target visible image data, construct a visibility consistency constraint model. The visibility consistency constraint model is used to ensure that the generated image and the target visible image are consistent at the visual perception level under normal observation conditions.

[0029] Step 103: Construct a hidden separability constraint model for each layer of hidden information data. The hidden separability constraint model is used to limit the hidden information of each layer to have distinguishable decoding features under preset recognition conditions.

[0030] Step 104: Under the condition that the visible consistency constraint model and each of the invisible separability constraint models are satisfied at the same time, perform joint optimization calculation on the target visible image data and the invisible information data of each layer to generate single printing data, and use the single printing data for the printing output of cultural and creative packaging.

[0031] As can be seen from the above process, this application constructs a visible consistency constraint model and a hidden separability constraint model, and performs joint optimization calculations on the target visible image and multi-layer hidden information under dual constraints to generate single printed data. This achieves the following: after the cultural and creative packaging is printed, it maintains a high degree of visual consistency with the original visible image under normal observation conditions, and reliably decodes the corresponding layer of hidden information layer by layer under different preset recognition conditions. This method effectively solves the traditional problem of embedding multi-layer hidden information in the highly artistic and complex texture background of cultural and creative packaging. It ensures both extremely high imperceptibility and aesthetics, significantly improves inter-layer distinguishability and detection robustness, and avoids interference and capacity loss caused by independent layer-by-layer embedding. It is suitable for interactive anti-counterfeiting and narrative enhancement of cultural and creative products such as tea, books, and intangible cultural heritage derivatives, and has outstanding practical value and innovation.

[0032] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments.

[0033] First, the above step 101, namely "acquiring target visible image data for cultural and creative packaging, and at least two layers of hidden information data to be embedded", will be described in detail with reference to the embodiments.

[0034] Visible image data refers to the main visual content that will ultimately be presented to consumers on the surface of cultural and creative packaging; that is, the complete design pattern visible to the naked eye on the packaging box, gift box, paper bag, or label. These images typically possess high artistic and cultural connotations, such as traditional ink wash landscapes, paper-cut patterns, embroidery designs, intangible cultural heritage elements, hand-drawn illustrations, or brand story illustrations. These images often feature complex textures, rich colors, and dense details, demanding an extremely high degree of imperceptibility in embedding implicit information.

[0035] The requirement of at least two layers of embedded hidden information refers to multiple sets of secret information hidden within the visible images, each layer existing independently and without interference. This hidden information can take the form of text, numerical sequences, binary codes, phrases, poems, artisan signatures, limited edition numbers, blessings, unlocking passwords, AR trigger markers, music clip indexes, etc. The requirement of at least two layers is to achieve a tiered or narrative-style interactive experience. For example, the first layer is readable under ordinary visible light for basic anti-counterfeiting verification; the second layer requires specific conditions (such as ultraviolet light, polarizing filters, or app filters) to unlock, providing deeper cultural stories, limited-edition benefits, or personalized content. This multi-layered design imbues cultural and creative packaging with a sense of ritual and collectible value through "layer-by-layer unveiling," distinguishing it from traditional single-layer anti-counterfeiting practices.

[0036] In practice, the target visible image data is usually input in the form of high-resolution digital files, such as PNG, TIFF, or PSD files in RGB or CMYK format, with a resolution no lower than printing requirements (e.g., 300 dpi or higher). The latent information data to be embedded is first encoded and converted into a bitstream or modulated signal to ensure sufficient robustness and distortion resistance during subsequent embedding.

[0037] The following describes in detail step 102, namely, "based on the target visible image data, construct a visible consistency constraint model, wherein the visible consistency constraint model is used to limit the generated image to be consistent with the target visible image at the visual perception level under normal observation conditions".

[0038] The core task of this step is to establish a set of mathematical or algorithmic constraints on the input target visible image data. These rules are specifically designed to ensure that, after embedding multiple layers of hidden information and generating the final printed image, when consumers observe the packaging with the naked eye under normal lighting, distance, and angle, the generated image is visually almost imperceptibly different from the original target visible image. In other words, the human eye cannot detect any blurring, color differences, texture distortion, or unusual marks on the packaging caused by the hidden information, thus ensuring that the aesthetics and cultural expression of the creative packaging are not compromised.

[0039] The process of constructing a visibility consistency constraint model typically begins with the perceptual feature decomposition of the target visible image data. Through multi-scale analysis or feature extraction methods relevant to the human visual system, the image is decomposed into structural components of different frequencies, such as large low-frequency outlines and fine high-frequency textures. Among these components, those most relevant to human visual stability are preferentially extracted, forming a set of structural features. This set reflects the perceptually sensitive regions of the image under different viewing scales, illumination intensities, and contrasts.

[0040] Based on the extracted set of structural features, a visible stability interval is further determined. This interval describes the tolerance limit of the human eye to local changes in an image under various normal observation conditions. For example, when observing under bright natural light, certain high-texture areas allow for larger pixel perturbations, while under low light or sidelight conditions, the perturbation threshold for the same area will be significantly tightened. Through statistical or modeling methods, these tolerance limits are quantified into a visible stability interval and mapped to specific constraint feature description parameters. These parameters can be in the form of upper limits for pixel value deviation, texture energy change thresholds, lower limits for perceptual quality scores, etc.

[0041] As an implementable approach, constructing a visibility consistency constraint model based on the target visible image data includes: performing perceptual feature decomposition on the target visible image data to extract a set of structural features related to human visual stability; determining a visibility stability interval that maintains visual perception stability under different observation scales and illumination conditions based on the set of structural features; mapping the visibility stability interval to constraint feature description parameters, and using the constraint feature description parameters to limit the range of image changes allowed during the embedding of latent information, thereby forming the visibility consistency constraint model.

[0042] Specifically, the first step in building the model is to perform perceptual feature decomposition on the target visible image data. This process draws on the principle that the human visual system has different sensitivities to different frequency components of an image. It typically employs multi-scale wavelet transform, pyramid decomposition, Fourier transform, or feature extraction layers of deep neural networks to break down the original image into low-frequency structural information and high-frequency detail texture. The low-frequency components correspond to the overall outline, color blocks, and large smooth areas of the image; these areas are more sensitive to subtle changes. The high-frequency components correspond to edges, textures, and fine patterns; these areas are more visually tolerant of a certain degree of disturbance. Through this decomposition, the system can identify which parts of the image are visually stable "safe zones" and which parts are prone to visible defects due to embedded information, thus providing a targeted basis for subsequent constraints.

[0043] Next, based on the extracted set of structural features, a visible stable interval is determined that maintains visual perception stability under different observation scales and lighting conditions. This interval is essentially a set of tolerance boundaries for local changes in an image. It considers various scenarios in which consumers observe packaging in real-world situations, such as browsing from a distance on a shelf, opening and handling the package at close range, indoor natural light, shopping mall spotlights, or outdoor sunlight. At different scales, the human eye's ability to perceive details varies; under different lighting conditions, the sensitivity to contrast and color shifts also differs. For example, in low-light environments, the human eye's threshold for perceiving changes in brightness increases, while its tolerance for color differences decreases under strong light. Through statistical analysis or calculations using human visual models, such as contrast sensitivity functions, visual saliency maps, or multi-scale structural similarity indices, the system defines a "stable interval" for each sub-region or frequency component of the image; that is, any modification within this interval will not cause a difference perceptible to the human eye.

[0044] Then, the visible stability interval is mapped to constraint feature description parameters. This mapping is the process of transforming qualitative tolerances at the perceptual level into computable quantitative indicators. Constraint feature description parameters can take various forms, such as pixel-level maximum allowable deviation, upper limit of texture energy variation, lower limit of perceptual quality score, and local contrast perturbation threshold. These parameters are not uniform global values, but are adaptively generated based on the features of different regions of the image. For example, the parameter thresholds are more lenient in high-texture regions and more stringent in low-texture smooth regions. This adaptive mapping ensures the adaptability of the constraint model to complex cultural patterns on creative packaging, avoiding the waste of capacity or visible distortion caused by a "one-size-fits-all" approach.

[0045] Ultimately, these constraint features define the range of image variations allowed during the latent information embedding process, thus forming a complete visibility consistency constraint model. In joint optimization or embedding computation, any modification to the image must be strictly controlled within the parameter-defined range. If the perturbation caused by the embedding exceeds this range, the system will force a pull back through a loss function or penalty term, thereby ensuring that the generated printed image is highly consistent with the original target visible image at the visual perception level under normal viewing conditions.

[0046] The following describes in detail step 103, namely, "constructing a hidden separability constraint model for each layer of hidden information data, wherein the hidden separability constraint model is used to limit the hidden information of each layer to have distinguishable decoding features under preset recognition conditions".

[0047] The construction of the cloaking separability constraint model is performed independently for each layer of cloaking information data, rather than processing all layers uniformly. This separate construction approach stems from the inherent differences between the multiple layers of information: each layer may correspond to different cultural meanings, unlocking difficulties, or privilege levels. For example, the first layer might be a basic anti-counterfeiting mark, the second layer a craftsman's note, and the third layer a limited-edition story sequel. For each layer, the system first establishes a corresponding set of preset recognition condition parameters. These conditions may include one or more of the following: illumination band, imaging angle, imaging resolution, image acquisition device type, or image preprocessing parameters. These are the "keys" that users actually use during the detection phase. Through these parameters, the model simulates the response characteristics of each layer of information under specific conditions. For example, under ultraviolet light, the first layer of texture responds strongly, while other layers maintain low responses.

[0048] Based on the set of recognition condition parameters, the system extracts hierarchical discriminative features for each layer. These features can be frequency domain statistics, color channel responses, texture microstructure perturbations, embedding vectors extracted by deep networks, etc. The extraction process ensures that the features are highly sensitive to the corresponding recognition conditions, while being relatively insensitive to other conditions, thus laying the foundation for subsequent separation. The discriminative features of all layers are projected into a unified discriminative feature space, which can be a high-dimensional vector space or a dimension-reduced embedding space. Within this space, the distribution of features for each layer is modeled to obtain the corresponding discriminative region for each layer. These regions represent the "territory" of the information of that layer in the feature space.

[0049] As an implementable approach, the step of constructing a stealth separability constraint model for each layer of stealth information data includes: establishing a corresponding set of identification condition parameters for each layer of stealth information data, and extracting hierarchical discrimination features of the stealth information based on the set of identification condition parameters; performing distribution modeling on the hierarchical discrimination features of each layer of stealth information within a unified discrimination feature space to obtain the discrimination region corresponding to each layer of stealth information; and, based on the interrelationship between the discrimination regions, limiting the minimum distinguishing interval of each layer of stealth information in the discrimination feature space, and mapping the minimum distinguishing interval to the constraint conditions of the stealth information embedding stage, thereby forming the stealth separability constraint model.

[0050] Specifically, for each layer, the system establishes a dedicated set of recognition condition parameters. These parameters define the most sensitive triggering methods for detecting information at that layer, including specific illumination bands, polarization angle ranges, imaging distance thresholds, filter types, and image preprocessing rules. Through these parameters, the model simulates and enhances the unique response characteristics of the layer's information under corresponding conditions, while suppressing its performance under other conditions.

[0051] Next, hierarchical discriminative features are extracted for each layer based on the set of recognition condition parameters. This extraction process is highly dependent on the condition parameters; for example, fluorescence response statistics may be extracted under ultraviolet light, texture directionality features under polarization filters, and frequency domain coefficients after color channel separation under App filters. These features are designed to be highly sensitive to the corresponding conditions of the layer, while being relatively insensitive to other conditions, thus providing a reliable basis for separation. The discriminative features of all layers are then projected into a unified discriminative feature space. This space can be a high-dimensional vector space, an embedding space, or a dimension-reduced subspace, providing a common coordinate system for cross-layer comparisons.

[0052] Within a unified discriminative feature space, the discriminative features of each layer are modeled using distributional methods, typically probability density estimation, Gaussian mixture models, kernel density estimation, or support vector domain description, to obtain the corresponding discriminative regions for each layer. These regions represent the "natural occupancy" or "confidence distribution" of the information of that layer in the feature space. The size, shape, and location of the regions directly reflect the response strength and stability of the layer's features under corresponding conditions. After modeling, the system analyzes the interrelationships between the discriminative regions of each layer, particularly indicators such as overlap, center distance, and boundary gap. Significant overlap indicates a high risk of inter-layer interference under certain conditions, which can easily lead to decoding confusion.

[0053] To address this issue, the model defines a minimum discriminative margin for each layer in the discriminative feature space based on the relationships between regions. This margin is a defined quantified threshold used to ensure that the separation between any two layer regions is not lower than this value. The minimum discriminative margin is not statically set but dynamically calculated based on the actual overlap and further optimized in subsequent adaptive adjustments. It is directly mapped to a constraint in the latent information embedding stage and participates in joint optimization calculations, such as penalizing region overlap or forcibly increasing the inter-class distance through the loss function, thereby pre-guaranteeing inter-layer discriminability when adjusting the embedding parameters.

[0054] Preferably, based on the interrelationships between the discrimination regions, the minimum distinguishing interval of each layer of hidden information in the discrimination feature space is defined, including: calculating the degree of overlap between each layer of discrimination regions; and adaptively adjusting the minimum distinguishing interval according to the degree of overlap based on a preset initial minimum distinguishing interval, wherein the adaptive adjustment includes: if the degree of overlap exceeds a preset threshold, increasing the minimum distinguishing interval of each relevant layer proportionally until the degree of overlap is lower than the threshold or reaches a preset upper limit; if the degree of overlap is lower than the threshold, reducing the minimum distinguishing interval proportionally to improve the embedding capacity, while ensuring that the preset minimum security interval requirement is not violated.

[0055] In the unified discriminative feature space, each layer of latent information forms a discriminative region after distribution modeling. These regions can be regarded as the "confidence territory" of that layer's information in the feature space. The degree of overlap is quantified by various indicators, such as the intersection-union ratio between regions, probability density integral overlap, inter-class distance margin deficiency, or Gaussian distribution overlap coefficient. A higher degree of overlap indicates that under certain preset recognition conditions, the discriminative feature responses of two or more layers are too similar, which can easily lead to cross-activation or misjudgment during decoding. This calculation result directly reflects the risk level of inter-layer interference under the current embedding parameters, providing an objective basis for subsequent adjustments.

[0056] The initial minimum distinguishing interval is a pre-set empirical value or a starting point calculated based on the number of layers and the dimension of the feature space, such as the total feature space scale divided by the number of layers, or a default safety distance obtained from preliminary embedding tests. This initial value ensures that even in the worst case, the regions of each layer will not completely overlap. Adaptive adjustment dynamically modifies this interval according to the actual degree of overlap, making it no longer a static parameter, but a variable threshold that can respond to specific image content and latent information characteristics. This dynamism is lacking in traditional multi-layer watermarking methods, which often rely on fixed thresholds or manual parameter tuning, making it difficult to adapt to the diversity of textures in cultural and creative packaging.

[0057] When the overlap exceeds a preset threshold, the system determines that the current separation is insufficient and increases the minimum distinguishing interval of each relevant layer proportionally. This increase can be a linear scaling up, an exponential scaling up, or a gradient increment based on the overlap, for example, the interval adjustment is proportional to the overlap ratio. After adjustment, the overlap is recalculated. If it still exceeds the limit, the system continues to iterate until the overlap falls below the threshold or the adjustment reaches the preset upper limit, to prevent excessive scaling from causing a sharp drop in embedding capacity. This iterative mechanism ensures that the separation is forcibly satisfied while avoiding the inefficiency of blind scaling.

[0058] When the overlap is below a preset threshold, the system determines that the current separation margin is sufficient. At this point, it allows a proportional reduction in the minimum distinguishing interval to free up more embedding space and increase the capacity or bit rate of the hidden information. This reduction operation is also strictly constrained and must be guaranteed to be no less than the preset minimum safety interval requirement. This safety lower limit is a baseline value set based on experience or theoretical analysis to prevent separation failure caused by printing distortion, noise, or acquisition errors.

[0059] By mapping the adjusted minimum discriminative margin back to the constraints of the embedding stage, this feature forms a complete closed-loop optimization path. In the joint optimization computation, the minimum discriminative margin, as part of the loss or penalty term, directly affects the selection of the embedding parameters. Optimization can only converge when the overlap of all layer discrimination regions is controlled within a safe range, thus ensuring that the generated single print data has reliable layer-by-layer separation capability during actual detection.

[0060] Through this constraint model, this invention achieves "condition-driven distinguishability": the preset recognition conditions are not only triggers during detection but also design guidelines for the embedding stage. Only by optimizing the embedding parameters under model constraints can the final printed image remain invisible under normal observation, while becoming clearly visible layer by layer under specific conditions. This mechanism is particularly suitable for the narrative needs of cultural and creative packaging. Users can unlock hidden content layer by layer by simply changing the light source, angle, or app filter, obtaining a complete cultural experience from basic verification to emotional resonance, while maintaining the reliability of anti-counterfeiting and the independence of multi-layer information. This feature effectively compensates for the shortcomings of traditional multi-layer watermarking, such as large inter-layer interference and unstable condition separation, and represents a significant technological advancement.

[0061] The following describes in detail step 104, namely, "under the condition that the visible consistency constraint model and each of the invisible separability constraint models are satisfied at the same time, joint optimization calculation is performed on the target visible image data and each layer of invisible information data to generate single printing data, and the single printing data is used for the printing output of cultural and creative packaging," with reference to the embodiments.

[0062] This step uses the previously constructed dual-constraint model as a strict optimization boundary. Through a unified calculation process, it simultaneously processes visible images and multiple layers of implicit information, ultimately outputting a single data file that can be directly used for printing. This process completely changes the traditional method of independent layer-by-layer operation for embedding multiple layers of implicit information, avoiding inter-layer parameter conflicts and accumulated errors, and achieving a high degree of unity in aesthetics, anti-counterfeiting, and interactive experience for cultural and creative packaging.

[0063] The core of joint optimization computation lies in simultaneously satisfying two constraint models. This means that the optimization objective is not simply to minimize a single loss, but to incorporate both the visibility consistency constraint and the latent separability constraint into the same optimization framework, forming a multi-objective collaborative constraint. During the optimization process, the system must ensure that any modification to the image does not violate the visibility consistency constraint, i.e., the generated image is visually consistent with the original target's visible image under normal observation conditions; simultaneously, it must not violate the latent separability constraint, i.e., under preset recognition conditions, the discriminative feature regions of each layer of latent information maintain sufficient distinguishing margins. This dual-constraint parallel constraint mechanism ensures the global optimality of the embedding operation, rather than the trade-offs caused by local optima.

[0064] The optimization computation is performed within a unified representation space. This space can be the pixel domain, a transform domain such as the wavelet domain or the discrete cosine domain, or an intermediate feature layer of a deep neural network. The visible image data of the target and the implicit information data from each layer are jointly mapped to this space, forming a joint representation structure. In this structure, all information components are treated as adjustable parameters. The optimization algorithm iteratively updates the representation state of these components, using methods such as gradient descent, surrogate gradients, or genetic algorithms. In each iteration, it simultaneously evaluates the visible consistency loss term and the implicit separability loss term. The visible consistency loss term focuses on perceptual quality, such as perceptual loss based on the human visual system and structural similarity metrics; the implicit separability loss term focuses on inter-layer separation, such as minimum distance constraints or mutual information minimization. The optimization process ends when the weighted sum of all loss terms falls below a preset convergence threshold, or when the number of iterations reaches its maximum, outputting the final single printed data.

[0065] As an implementable approach, the representation state of each information component in the joint representation structure is updated iteratively, including: iteratively optimizing the joint representation structure using gradient descent; in each iteration step, simultaneously calculating a visible consistency loss term and a hidden separability loss term; the visible consistency loss term includes at least a weighted combination of a perceptual loss function and a structural similarity measure; the hidden separability loss term includes a weighted combination of inter-layer minimum distance constraints and mutual information minimization constraints in the discriminative feature space; when the weighted sum of the visible consistency loss term and the hidden separability loss term is lower than a preset convergence threshold, or when the maximum number of iterations is reached, the optimization is determined to have converged and a single printed data is output.

[0066] Specifically, iteratively updating the representation state of each information component in the joint representation structure is the main action of the entire optimization process. The joint representation structure is an overall framework that unifies the mapping of the target's visible image data and the latent information data of each layer. It can be a pixel grid, a transform domain coefficient array, or an intermediate activation map of a neural network. In this structure, each information component corresponds to an adjustable variable, such as a pixel value, coefficient amplitude, or feature vector. The optimization process continuously adjusts the values ​​of these variables to ensure that the final generated image closely approximates the original visible image while allowing the discriminative features of each latent information layer to be clearly separated under corresponding conditions.

[0067] Iterative optimization of the joint representation structure using gradient descent is the current mainstream approach. This method gradually reduces the total loss by calculating the gradient of the total loss with respect to each variable and updating the variable values ​​in small steps along the inverse direction of the gradient. Gradient descent is advantageous because it is efficient, differentiable, and capable of handling large-scale parameter optimization. In each iteration, the system simultaneously calculates the visible consistency loss term and the latent separability loss term. These two losses correspond to two major constraint models: the visible consistency loss term measures the difference between the generated image and the original image in human visual perception, while the latent separability loss term measures the degree of separation of latent information in the discriminative feature space. By calculating them simultaneously, the optimizer can balance the two, avoiding optimizing one at the expense of the other.

[0068] It is evident that the consistency loss term comprises at least a weighted combination of a perceptual loss function and a structural similarity measure. This combination ensures that the generated image appears highly similar to the original image to the human eye. The perceptual loss function focuses on high-level semantic and textural similarity, capturing the human eye's overall perception of the content; the structural similarity measure, on the other hand, focuses on evaluating the degree of preservation of brightness, contrast, and structural information. The weighted combination of these two components can be adjusted according to the specific needs of cultural and creative packaging. For example, increasing the weight of the structural similarity measure in high-texture areas can protect the integrity of intricate cultural patterns.

[0069] The latent separability loss term comprises a weighted combination of inter-layer minimum distance constraints and mutual information minimization constraints in the discriminative feature space. This combination forces the discriminative regions of each layer's latent information to remain sufficiently separated in the feature space. The inter-layer minimum distance constraint increases the inter-class margin by penalizing regions with excessively close center distances or overlapping boundaries, ensuring that when a feature of one layer is prominent under corresponding recognition conditions, the responses of other layers are suppressed. The mutual information minimization constraint reduces inter-layer statistical dependence, making the features of each layer more independent. This combination ensures both hard separation and enhanced statistical independence, thus enabling reliable decoding even under printing distortion or acquisition errors.

[0070] When the weighted sum of the visible consistency loss term and the implicit separability loss term falls below a preset convergence threshold, or when the number of iterations reaches its maximum value, the system determines that the optimization has converged and outputs a single print data set. This convergence determination mechanism balances computational efficiency and optimization quality: a threshold that is too low may lead to over-optimization and waste time, while a threshold that is too high may result in failure to reach the optimal value; the maximum number of iterations serves as a safety upper limit to avoid getting trapped in local minima or infinite loops. The final output of the single print data set is a complete print file, directly used for the production of cultural and creative packaging, ensuring that the packaging perfectly presents the original design under normal observation, while the hidden multi-layered information can be reliably read layer by layer under preset conditions.

[0071] The generated single print data is a complete digital file that can be directly fed into the printing press, typically image data in CMYK four-color channels or spot color channels. It contains all the visual information of the original visible image, while also implicitly containing optimized, embedded layers of hidden information. This hidden information is perfectly concealed under normal lighting conditions and can only be effectively activated and decoded under specific preset recognition conditions. When this single print data is used for the printing output of cultural and creative packaging, the packaging surface presents an aesthetically pleasing effect that is completely consistent with the original design, while the hidden layers of information provide a solid foundation for subsequent interactive experiences.

[0072] According to another embodiment, a multi-layer invisible information detection system for cultural and creative packaging is provided. Figure 2 This diagram illustrates a schematic block diagram of a multi-layer invisible information detection system for cultural and creative packaging according to one embodiment. The system is used to detect the printed image formed after a single print data generated by a multi-layer invisible information printing method for cultural and creative packaging according to the above embodiment is printed onto the packaging, such as... Figure 2 As shown, the system 200 includes:

[0073] The data acquisition module 201 is used to acquire the detection image data of the printed image.

[0074] The perception condition modeling module 202 is used to model the perception feature space corresponding to the detected image data according to the preset recognition conditions.

[0075] The hierarchical detection control module 203 is used to determine the current level of hidden information to be detected based on the perceived feature space and the pre-established hierarchical detection rules.

[0076] The stealth information decoding module 204 is used to perform targeted decoding calculations on the detected image data at the determined stealth information level to obtain candidate stealth information at the corresponding level.

[0077] The consistency and confidence determination module 205 is used to verify the consistency between the candidate hidden information and the hidden information constraint features generated in the printing stage, and to calculate the corresponding decoding confidence.

[0078] The detection result generation module 206 is used to determine the detection results of each layer of hidden information based on the consistency and confidence level.

[0079] As can be seen from the above system, by constructing a multi-layered hidden information detection system specifically for cultural and creative packaging, reliable, layer-by-layer decoding of printed images formed after printing from single printing data generated by the method of claim 1 can be achieved. The system collects detection image data under preset recognition conditions and reproduces the perceptual feature space consistent with the embedding stage through a perceptual condition modeling module; the layer detection control module intelligently determines the current activation layer based on pre-established layer detection rules; the hidden information decoding module performs targeted decoding to obtain candidate information; the consistency and confidence judgment module rigorously verifies the candidate results and quantifies the confidence level; and the final detection result generation module filters, weights, fuses, and visualizes the information from each layer, while recording a complete log. This system effectively solves the decoding problems caused by printing distortion, acquisition errors, and inter-layer interference, ensuring clear layer-by-layer reading of hidden information under different lighting bands, polarization angles, or App filters. It also provides high-confidence, traceable detection results, significantly improving the interactive anti-counterfeiting experience and user trust in cultural and creative packaging, demonstrating strong practicality and robustness.

[0080] The data acquisition module 201 will be described in detail below with reference to the embodiments.

[0081] The main task of the data acquisition module is to capture and detect image data from the actually printed cultural and creative packaging. The printed image is the final visual content formed on the surface of the packaging after the single printed data generated according to the method given in this application is output by the printing press. This printed image surface carries all the appearance features of the original target visible image, while implicitly containing multiple layers of hidden information that have been jointly optimized and embedded. The acquisition module converts this information into digital form through hardware devices so that subsequent modules can perform perceptual modeling, hierarchical detection, and decoding processing.

[0082] In practical applications, data acquisition modules are typically integrated into mobile terminals or dedicated inspection equipment, primarily relying on cameras or image sensors to complete the data acquisition. When using the device, the user places the packaging at an appropriate distance and angle, and the device initiates recording under ambient light or auxiliary lighting. The acquired raw image data includes information such as the color, texture, lighting, and possible noise and distortion of the printed image. This raw data is the direct input to the inspection process and must reflect the true state of the printed surface as completely and clearly as possible.

[0083] The data acquisition process needs to adapt to various real-world usage scenarios. For example, when shooting under different lighting conditions, the module automatically adjusts exposure time, gain, or white balance parameters according to preset recognition conditions to ensure the stability of image brightness, contrast, and color. Under certain specific recognition conditions, such as when ultraviolet or polarized light assistance is required, the acquisition module will work with external light sources or filter components to ensure the capture of effective response information in the corresponding wavelength or polarization direction. In addition, to reduce interference from hand shake, perspective distortion, and other factors, the acquisition module typically performs preliminary image stabilization and geometric correction in real time, providing high-quality raw materials for subsequent perception condition modeling.

[0084] The acquired detection image data is the foundation for all subsequent processing, directly affecting the accuracy and confidence level of layer detection. If the acquired image quality is too low or fails to match the preset recognition conditions, subsequent modules may fail to correctly activate the discriminative features of the corresponding layer, or even lead to decoding failure. Therefore, the data acquisition module is not just a simple photo-taking tool, but the first guarantee of the robustness and user experience of the entire detection system. By reliably and with high fidelity acquiring the digital representation of printed images, it provides a solid foundation for perceptual condition modeling, layer control, and layer-by-layer decoding, enabling the multiple layers of hidden information concealed in cultural and creative packaging to be accurately and progressively activated in real-world scenarios.

[0085] The perception condition modeling module 202 will be described in detail below with reference to the embodiments.

[0086] The perceptual condition modeling module 202 is responsible for transforming the acquired raw detection image data into a perceptual feature space that highly matches the embedding stage, thereby providing an accurate and consistent feature foundation for subsequent hierarchical detection and layer-by-layer decoding. This module ensures that the entire detection process can "reproduce" the perceptual constraints considered during printing, achieving a reliable closed loop from physical printing to digital decoding.

[0087] The perceptual condition modeling module is configured to: perform feature extraction processing on the detected image data under the preset recognition conditions, which is adapted to or corresponds to the perceptual feature decomposition method in the visible consistency constraint model of the multi-layer invisible information printing method for cultural and creative packaging given in this application, so as to construct a perceptual feature space for hierarchical detection.

[0088] The core function of the perceptual condition modeling module is to perform targeted feature space transformation on the input detection image data based on the preset recognition conditions actually matched during the current detection. These preset recognition conditions are trigger parameters defined during the embedding stage, such as specific illumination bands, polarization angles, imaging distances, filter types, or image preprocessing rules. These conditions are identified by the user or automatically by the device during detection. Once a match is successful, the module loads the corresponding perceptual transformation rules and performs condition-specific processing on the detection image. This processing is not simple image enhancement, but rather simulates the perceptual feature decomposition method used by the visible consistency constraint model during embedding, thereby constructing the same perceptual feature space as the printing optimization stage.

[0089] The modeling process begins by applying condition-specific preprocessing steps to the detected image data. For example, if the preset recognition condition is ultraviolet light with a 45-degree polarization filter, the module will first separate the ultraviolet response channel and apply a polarization decoding algorithm; if the condition is an App style transfer filter, it will perform the corresponding color space conversion or frequency domain filtering. These preprocessing steps ensure that the detected image is as close as possible to the simulated state at the feature level, avoiding feature drift caused by deviations in acquisition conditions. The processed image then enters the perceptual feature extraction stage, typically employing the same multi-scale decomposition methods as in the visibility consistency constraint model, such as wavelet transform, pyramid decomposition, or intermediate layer activation of a pre-trained network, thereby generating a set of structured, hierarchical perceptual features.

[0090] The generated perceptual feature space is a multi-dimensional vector space or feature map set that retains key information of the image in the dimensions of human visual perception while incorporating the specific response characteristics of preset recognition conditions. Within this space, the discriminative features of each layer of hidden information can be projected and compared more accurately. The module's output is directly supplied to the hierarchical detection control module for matching with pre-established hierarchical detection rules, thereby determining whether the current conditions can activate a specific layer. If the perceptual feature space modeling is inaccurate, subsequent hierarchical activation and decoding will deviate, potentially leading to misreading or omission of hidden information.

[0091] Through this module, the present invention achieves a deep correspondence between the detection system and the printing method: the dual-constraint model in the embedding stage fully considers the preset recognition conditions during optimization, while the perception condition modeling in the detection stage reproduces the feature responses under these conditions in reverse. This symmetrical design greatly improves the robustness and accuracy of the system. Especially in the actual use scenario of cultural and creative packaging, users may face problems such as changes in lighting, hand tremors, or differences in devices. The perception condition modeling module effectively compensates for these variables through adaptive modeling, ensuring that multiple layers of implicit information can be stably and progressively awakened under different triggering conditions, bringing users a reliable and ritualistic interactive experience.

[0092] The hierarchical detection and control module 203 will be described in detail below with reference to the embodiments.

[0093] The hierarchical detection control module starts with the perceptual feature space. This space, generated by the perceptual condition modeling module based on current preset recognition conditions, contains structured feature representations of the detected image in the dimensions of human visual perception and under specific triggering conditions. These features directly correspond to the discriminative feature space of the embedding stage, thus possessing high comparability. The module takes the current perceptual feature space as input and matches it with pre-established hierarchical detection rules. These rules are the product of joint computation convergence during the printing optimization stage and typically include information such as the classification boundaries, center coordinates, minimum discrimination interval threshold, activation confidence lower bound, and layer-by-layer priority order of each layer's discriminative region. These rules are essentially a "mirror image" or "decision mapping" of the embedding constraint model, ensuring consistency between the detection process and the embedding process at the feature level.

[0094] Determining the current hidden information level to be detected is an intelligent matching and decision-making process. The module first projects key feature vectors or feature maps from the perceptual feature space onto a unified discriminative feature space, and then calculates the matching degree between the current feature and the preset discriminative region of each layer. This matching degree can be quantified based on Euclidean distance, cosine similarity, Mahalanobis distance, probability score, or support vector classification. For each layer, if the current feature falls within the discriminative region of that layer and is sufficiently close to the center, and the matching degree is higher than a preset activation threshold, then that layer is determined to be the current detection level. Simultaneously, the module considers the risk of inter-layer interference: if the current feature is close to multiple regions simultaneously, it prioritizes the layer with the highest confidence or priority, and may suppress the activation of other layers to avoid cross-reading.

[0095] The hierarchical detection control module typically employs a layer-by-layer activation strategy, starting with the most easily triggered layer, such as those corresponding to visible light or default conditions. If the current conditions match the rules of that layer, the corresponding decoding path is activated directly; otherwise, the module attempts the next layer sequentially or prompts the user to adjust the recognition conditions. This strategy significantly improves detection efficiency, especially in real-world scenarios where users are using handheld devices and ambient light changes frequently, enabling rapid response and avoiding unnecessary calculations. The module also dynamically adjusts subsequent processing based on the matching results. For example, when a high-confidence layer is detected, the remaining layers can be skipped directly, or confidence sorting and fusion preparation can be performed when multiple layers are activated simultaneously.

[0096] The stealth information decoding module 204 will be described in detail below with reference to the embodiments.

[0097] The stealth information decoding module 204 is activated after the layer detection control module confirms the current layer to be detected. Based on the specified layer characteristics, it performs dedicated reverse calculations on the detected image data to output the candidate stealth information corresponding to that layer. This module is the key transformation link from "detection" to "reading" in the entire system, ensuring that multi-layer stealth information can be accurately restored under the corresponding trigger conditions.

[0098] The decoding module's operation is highly dependent on the layer determination results output by the front-end module. Once the layer detection and control module determines that the current conditions correspond to a specific layer—for example, the first layer is activated under visible light or the second layer is activated under ultraviolet light with a polarizing filter—the decoding module loads the layer's dedicated decoding path. These paths are inverse mappings designed during the printing optimization stage, including specific feature extraction rules, inverse transform algorithms, error correction mechanisms, and bitstream restoration logic. This targeted approach is reflected in the fact that the decoding calculations are entirely customized based on the preset recognition conditions and discrimination features of that layer, avoiding the inefficiency and interference of uniformly processing all layers.

[0099] Performing targeted decoding calculations is the core operation of the module. For different layers, the calculation process may involve various techniques. For example, for layers based on frequency domain embedding, the module performs inverse discrete cosine transform or inverse wavelet transform to extract perturbation signals from specific subbands of the detected image; for layers based on texture microstructure or color channel separation, directional filtering, channel subtraction, or statistical demodulation algorithms may be applied. These calculations directly affect the detected image data or its perceptual feature representation, gradually stripping away visible image interference and separating the original perturbation patterns of the latent information at that layer. Error-correcting code decoding, bit synchronization, and noise suppression steps are also incorporated into the calculation process to address the effects of printing distortion, acquisition noise, or illumination deviations.

[0100] After decoding is complete, the module outputs candidate latent information for the corresponding level. This candidate information is typically a bitstream, text sequence, binary code, or encoded data block, such as the restored artisan's signature poem, limited edition number string, AR trigger identifier, or blessing content. This candidate information is not the final result, but rather a preliminary extraction with a certain level of confidence, requiring verification by the subsequent consistency and confidence assessment module. Only candidate information that passes verification is considered a valid decoding result and passed to the detection result generation module for visualization.

[0101] This module enables precise and condition-specific reading of multi-layered hidden information. It ensures that each layer of information receives the most suitable decoding strategy upon activation, avoiding inter-layer crosstalk or low success rate issues caused by universal decoders.

[0102] The consistency and confidence determination module 205 will be described in detail below with reference to the embodiments.

[0103] The consistency and confidence level determination module is used to verify the consistency between the candidate hidden information and the constraint features of the hidden information generated during the printing stage, and to calculate the corresponding decoding confidence level. This technical feature is the final quality control and reliability assessment step in the detection system. It is responsible for strictly verifying the candidate results output by the hidden information decoding module to ensure that the decoded information is true and reliable, and is not a false result caused by noise interference or false activation. At the same time, it quantifies the credibility of the entire decoding process, providing a scientific basis for the final output. The existence of this module greatly improves the system's ability to prevent false judgments and user trust, especially in the actual use scenario of cultural and creative packaging, where it can effectively distinguish between real hidden content and random interference.

[0104] Consistency verification is one of the core actions of the module. Candidate latent information is a preliminary reconstruction result obtained from the detected image data through targeted decoding calculations, and may contain text, numerical sequences, bitstreams, or other encoded data. This candidate information needs to be compared with the latent information constraint features generated during the printing stage. These constraint features are reference standards that have been fixed during embedding optimization, such as the original feature summary of the latent information before embedding, checksum, hash value, expected perturbation pattern of the embedding position, center coordinates or distribution parameters in the discriminant feature space, etc. The verification process judges consistency by calculating the similarity or matching degree between the candidate information and these reference constraint features, such as using Hamming distance, correlation coefficient, cosine similarity of feature vectors, or checksum comparison. If the candidate information and the constraint features are highly consistent, it is judged as consistent; if the deviation is too large, it is judged as inconsistent, which may be caused by acquisition error, printing distortion, or incorrect level activation.

[0105] Confidence calculation is the quantitative output of the verification result. Even if candidate information passes consistency verification, a 100% perfect match is impossible because factors such as lighting variations, lens distortion, and noise interference always exist in actual detection. The module calculates the decoding confidence score by integrating multiple dimensions, such as matching similarity score, signal-to-noise ratio, error correction code correction ratio, feature space distance margin, and statistical stability of multiple samplings. These factors are weighted and fused into a confidence score, usually expressed as a percentage or a probability value between 0 and 1. The higher the confidence score, the more reliable the decoding result of the hidden information at that layer; if the confidence score is lower than a preset threshold, even if the consistency verification passes, it may be filtered out by subsequent modules to avoid presenting low-confidence content to the user.

[0106] The detection result generation module 206 will be described in detail below with reference to the embodiments.

[0107] The detection result generation module operates primarily based on consistency verification and confidence scores. Consistency verification indicates whether candidate implicit information highly matches the constraints of the printing stage, while the confidence score quantifies the reliability of this match. The module first filters candidate information at all levels: candidates with confidence scores below a preset threshold are directly eliminated to avoid presenting users with unreliable or potentially noisy results; candidates that pass verification and have high confidence scores are retained as valid detection information. This filtering process ensures high accuracy and security in the output results, preventing misleading users or reducing product trust.

[0108] For the hidden information at each layer that passes the screening, the module performs weighted fusion based on confidence level to generate a comprehensive detection result. This fusion is not a simple superposition, but considers the priority of different layers, confidence weights, and logical relationships between layers. For example, if the first layer has extremely high confidence while the second layer has low confidence, the module may prioritize presenting the first layer's content and refer to the second layer with a lower weight; if multiple layers are activated simultaneously and all have high confidence, they are fused into a complete multi-layered narrative sequence. The fused result is mapped to a multi-layered visual output, including the specific content of each layer of hidden information, its display location, corresponding area annotations on the packaging image, decoding status markers, and confidence value indicators. This visual presentation allows users to intuitively see the location and reliability of the hidden information, such as overlaying text, icons, or animations on the app interface to enhance interactivity and immersive experience.

[0109] Simultaneously, the module records a complete decoding log for each output message, including the currently used preset recognition conditions, decoding trigger time, confidence score, verification success details, and possible anomaly prompts. These logs form a traceable detection record, facilitating users' review of the historical decoding process and providing data support for subsequent system maintenance, problem diagnosis, or anti-counterfeiting auditing. Ultimately, the module presents both the visual output and the detection logs to the user or passes them to subsequent system calls, ensuring that the detection results are both intuitive and easy to understand, while also possessing complete technical traceability.

[0110] To further illustrate the technical effects of this application, a specific implementation method and the test results of this implementation method are given below.

[0111] In this embodiment, a typical landscape painting in the traditional Chinese style used for cultural and creative packaging is selected as the target visible image data, with a resolution of 2048×2048 pixels and using the RGB color space. The embedded hidden information includes two layers: the first layer is the craftsman's signature poem "Mountains and rivers convey meaning, ink charm endures" (the bitstream length is 128 bits after conversion to ASCII encoding), and the second layer is the limited edition blessing "May you enjoy longevity and happiness, and may fate bring you together" (the bitstream length is 160 bits).

[0112] First, a visibility consistency constraint model is constructed. The target visible image is decomposed into three scales using wavelet decomposition to extract low-frequency structural and high-frequency texture feature sets. Based on the contrast sensitivity function of the human visual system and multi-scale structural similarity indices, the visible stability intervals under different observation scales (1:1, 1:2, 1:4) and illumination conditions (standard illumination 500 lux, low illumination 100 lux) are determined and mapped to constraint feature description parameters, including a maximum permissible deviation threshold of 5 for pixel values ​​and an upper limit of 8% for local texture energy variation.

[0113] Next, a hidden separability constraint model is constructed. The preset recognition conditions are: the first layer corresponds to "visible light + normal shooting angle (vertical ±15°)"; the second layer corresponds to "ultraviolet light band 365nm + 45° polarization filter". A set of recognition condition parameters is established for each layer, and hierarchical discriminant features are extracted (the first layer uses DWT low-frequency coefficient statistics, and the second layer uses polarization response directionality features). Within a 128-dimensional unified discriminant feature space, a Gaussian mixture model is used to model the distribution of features at each layer, obtaining the discriminant region. The initial minimum discriminant interval is set to 0.1, and the overlap between regions is calculated to be 0.38 (IoU index). When the overlap exceeds the threshold of 0.2, the interval is increased proportionally by a factor of 1.6. After three iterations, the overlap decreases to 0.04, and the adjusted interval is 0.28. Once the overlap falls below the threshold, it is not further reduced to ensure a safety margin. The final minimum discriminant interval is mapped to embedded constraints.

[0114] A joint representation structure is constructed in a unified pixel domain, and visible consistency constraints and latent separability constraints are applied. The Adam optimizer is used for iterative optimization with a learning rate of 0.001. The visible consistency loss term is a weighted combination of the perceptual loss function LPIPS and the multi-scale structural similarity function MS-SSIM (weights 0.6:0.4); the latent separability loss term is a weighted combination of the inter-layer minimum distance constraint (hingeloss, margin=0.28) and the mutual information minimization constraint (weights 0.7:0.3). After 2000 iterations, the weighted total loss is below 0.012, indicating convergence, and a single print dataset (CMYK four-color print file) is output.

[0115] A single printout of data was fed into an inkjet printer, outputting an actual packaging sample on coated paper. A mobile app was used to perform detection under two preset recognition conditions: the first layer of poetry was successfully decoded in visible light mode with a confidence level of 96.8%; after switching to ultraviolet light with a 45° polarizing filter, the second layer of blessings was successfully decoded with a confidence level of 94.2%. There were no cross-reading errors between the two layers of information, and the average decoding time was 1.8 seconds.

[0116] To verify the effectiveness of this solution, a comparative experiment was conducted on the same cultural and creative landscape painting background. The traditional layer-by-layer embedding method (layer-by-layer DCT watermarking + empirical strength adjustment) achieved an average visual PSNR of 32.1dB and an inter-layer error rate of 18.7% under the same embedding capacity. After joint optimization, the visual PSNR increased to 38.4dB, and the inter-layer error rate decreased to 2.9%. In printing distortion simulation (adding Gaussian noise σ=5, rotation ±5°, JPEG compression quality 70), the decoding success rate of the first layer of this solution was 98.5%, and that of the second layer was 96.3%, while the traditional method decreased to 81.2% and 74.6%, respectively. In 50 user tests on actual cultural and creative packaging samples, the activation success rate of the first layer was 100% under normal lighting conditions, and the activation success rate of the second layer was 98% under ultraviolet + polarization conditions. The test results show that this solution significantly improves the robustness and inter-layer separation of multi-layer hidden information while maintaining high visual quality, making it suitable for the actual production and application of cultural and creative packaging.

[0117] The methods provided in this application can be applied to various application scenarios, including but not limited to: In holiday gifts such as Mid-Autumn Festival mooncake boxes or Dragon Boat Festival sachets, multiple layers of blessings and benefits information can be embedded: one layer is a general holiday greeting, one layer is a lottery code or membership benefit code, and one layer is a personalized voice blessing. Users read these layers one by one under specific conditions, combined with social sharing functions, to promote brand communication and user stickiness; In intangible cultural heritage derivatives such as embroidered bookmarks, paper-cut postcards, or ceramic derivative packaging, multiple layers of cultural narratives can be hidden: one layer is a basic introduction to the heritage, one layer is the artisan's signature or hand-drawn draft, and one layer is an AR animation trigger code. Users unlock these layers one by one through different recognition conditions, enhancing the immersive experience and educational value of cultural dissemination.

[0118] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0119] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. Components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0120] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0121] In addition, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in any of the foregoing method embodiments.

[0122] And an electronic device comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method described in any of the foregoing method embodiments.

[0123] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the foregoing method embodiments.

[0124] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for printing multi-layer invisible information for cultural and creative packaging, characterized in that, The method includes: Acquire the target visible image data for cultural and creative packaging, as well as at least two layers of implicit information data to be embedded; Based on the target visible image data, a visibility consistency constraint model is constructed. The visibility consistency constraint model is used to ensure that the generated image and the target visible image are consistent at the visual perception level under normal observation conditions. For each layer of hidden information data, a hidden separability constraint model is constructed. The hidden separability constraint model is used to limit the hidden information of each layer to have distinguishable decoding features under preset recognition conditions. Under the condition that both the visible consistency constraint model and each of the invisible separability constraint models are satisfied, joint optimization calculation is performed on the target visible image data and the invisible information data of each layer to generate single printing data, and the single printing data is used for the printing output of cultural and creative packaging.

2. The method according to claim 1, characterized in that, The construction of a visibility consistency constraint model based on the target visible image data includes: Perceptual feature decomposition is performed on the target visible image data to extract a set of structural features related to human visual stability; Based on the set of structural features, a visible stable range that maintains visual perception stability under different observation scales and lighting conditions is determined. The visible stability interval is mapped to constraint feature description parameters, and the range of image changes allowed to occur during the embedding of latent information is limited by the constraint feature description parameters, thereby forming the visible consistency constraint model.

3. The method according to claim 1, characterized in that, The preset recognition conditions include at least one or more of the following: illumination band, imaging angle, imaging resolution, image acquisition device type, or image preprocessing parameters.

4. The method according to claim 1, characterized in that, The aforementioned method constructs a hidden separability constraint model for each layer of hidden information data, including: For each layer of hidden information data, a corresponding set of identification condition parameters is established, and the hierarchical discrimination features of the hidden information are extracted based on the set of identification condition parameters. Within a unified discriminant feature space, the hierarchical discriminant features of each layer of latent information are distributed and modeled to obtain the discriminant regions corresponding to each layer of latent information. Based on the interrelationships between the discrimination regions, the minimum distinguishing interval of each layer of latent information in the discrimination feature space is defined, and the minimum distinguishing interval is mapped as a constraint condition in the latent information embedding stage, thereby forming the latent separability constraint model.

5. The method according to claim 4, characterized in that, The step of defining the minimum distinguishing interval of each layer of hidden information in the discriminative feature space based on the interrelationship between the discriminative regions includes: Calculate the degree of overlap between the discrimination regions of each layer; Based on a preset initial minimum discrimination interval, the minimum discrimination interval is adaptively adjusted according to the degree of overlap. The adaptive adjustment includes: If the degree of overlap exceeds a preset threshold, the minimum separation interval of each related layer is increased proportionally until the degree of overlap is lower than the threshold or reaches a preset upper limit; if the degree of overlap is lower than the threshold, the minimum separation interval is reduced proportionally to improve the embedding capacity, while ensuring that the preset minimum security interval requirement is not violated.

6. The method according to claim 1, characterized in that, The step of performing joint optimization calculations on the target visible image data and the hidden information data of each layer to generate single printing data includes: Within a unified data representation space, a joint representation structure is constructed to describe the target's visible image data and the hidden information data at each layer; Based on the visible consistency constraint model and each of the hidden separability constraint models, a collaborative constraint is applied to the joint representation structure to form a multi-constraint-restricted generation condition; Under the given generation conditions, the representation state of each information component in the joint representation structure is updated iteratively, and the corresponding single printed data is output when the convergence judgment conditions of each constraint model are met.

7. The method according to claim 6, characterized in that, The iterative update of the representation state of each information component in the joint representation structure includes: The joint representation structure is iteratively optimized using gradient descent. In each iteration step, both the visible consistency loss term and the implicit separability loss term are calculated simultaneously. The visible consistency loss term includes at least a weighted combination of the perceptual loss function and the structural similarity measure; The hidden separability loss term includes a weighted combination of inter-layer minimum distance constraints and mutual information minimization constraints in the discriminant feature space; When the weighted sum of the visible consistency loss term and the implicit separability loss term is lower than the preset convergence threshold, or when the maximum number of iterations is reached, optimization convergence is determined and a single printed data is output.

8. A multi-layer invisible information detection system for cultural and creative packaging, characterized in that, The system is used to detect the printed image formed after the single print data generated by the method of claim 1 is printed on cultural and creative packaging. The system includes: The data acquisition module is used to acquire the detection image data of the printed image; The perception condition modeling module is used to model the perception feature space corresponding to the detected image data according to the preset recognition conditions. The hierarchical detection control module is used to determine the current level of hidden information to be detected based on the perceived feature space and the pre-established hierarchical detection rules. The stealth information decoding module is used to perform targeted decoding calculations on the detected image data at the determined stealth information level to obtain candidate stealth information at the corresponding level. The consistency and confidence determination module is used to verify the consistency between the candidate hidden information and the hidden information constraint features generated in the printing stage, and to calculate the corresponding decoding confidence. The detection result generation module is used to determine the detection results of each layer of hidden information based on the consistency and confidence level.

9. The multi-layer invisible information detection system for cultural and creative packaging according to claim 8, characterized in that, The perception condition modeling module is configured as follows: Under the preset recognition conditions, feature extraction processing is performed on the detected image data in accordance with or corresponding to the perceptual feature decomposition method in the visible consistency constraint model of claim 2, so as to construct a perceptual feature space for hierarchical detection.

10. The system according to claim 8, characterized in that, The step of determining the detection results of each layer of hidden information based on the consistency and confidence level includes: The hidden information at each layer and its corresponding decoding confidence level are filtered out, and candidate information with a confidence level lower than a preset threshold is removed. The remaining hidden information from each layer is weighted and fused according to its confidence level to generate a comprehensive detection result; The comprehensive detection results are mapped into multi-layered visualization output, including the display position, decoding status and confidence index of each layer of hidden information; For each output message, the decoding conditions, decoding time, and confidence level are recorded to form a traceable detection log; The visualization output and detection logs are output simultaneously for use by users or subsequent systems.

Images